Industries / Media, arts and recreation / Professional sports
Player performance and health
This work is training, tracking and protecting athletes so they can keep playing: measuring workloads, testing exercises that prevent injury, spotting and managing head injuries, policing drugs, and counting injuries across a league.
6
Milestones achieved, from 1999 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1999
A world anti-doping agency
After a doping scandal in cycling in 1998, the International Olympic Committee (IOC) called a world conference on doping in Lausanne in February 1999. It led to the World Anti-Doping Agency (WADA), set up on 10 November 1999 and run and funded equally by sport and governments. WADA's single rulebook, the World Anti-Doping Code, was approved in Copenhagen in 2003 and came into force in 2004, and governments backed it through a UNESCO convention adopted on 19 October 2005, in force from 1 February 2007.
Credit
the IOC and the world's governments; WADA
2005
CTE found in NFL player
Bennet Omalu, a forensic pathologist, examined the brain of a retired NFL player who died about 12 years after he stopped playing. In the journal Neurosurgery in July 2005 he and his co-authors reported chronic traumatic encephalopathy (CTE), a brain disease linked to repeated blows to the head. Members of the NFL's own brain-injury committee asked the journal to retract the paper, and it was not retracted. In March 2016 an NFL executive told a congressional roundtable that the link between football and CTE was "certainly yes", and in 2017 a Boston University team found CTE in 110 of 111 brains donated by former NFL players, a group it called a convenience sample, meaning not chosen at random.
Lessonthe league's own committee tried to have the finding withdrawn, and outside researchers and then the league itself later backed it.
pubmed.ncbi.nlm.nih.govucs.orgnews.wfsu.orgpubmed.ncbi.nlm.nih.gov
Credit
Bennet Omalu, with DeKosky, Minster, Kamboh, Hamilton and Wecht (University of Pittsburgh); outside confirmation by Mez, McKee and colleagues (Boston University)
2008
Warm-up cuts injuries in trial
In a cluster randomized trial, whole clubs rather than single players are assigned at random. One such trial followed 1,892 girls aged 13 to 17 at 125 Norwegian clubs and tested a warm-up of about 20 minutes that became FIFA's "11+". Players who did it had 32% fewer injuries overall (rate ratio 0.68) and 45% fewer severe injuries (0.55), though the drop in leg injuries alone was not statistically significant. A 2011 Danish trial in 50 professional and amateur men's teams found the Nordic hamstring exercise cut acute hamstring injuries from 13.1 to 3.8 per 100 player-seasons, and a 2017 outside analysis of trials confirmed FIFA 11+ cut injuries (rate ratio 0.61), though all those trials were in recreational and sub-elite football.
Lessona 2015 survey of 50 elite teams found that 125 of 150 club-seasons (83.3%) did not complete the Nordic program, which the authors called too low to expect any effect on hamstring injury rates.
pubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.gov
Credit
Soligard, Bahr, Andersen and colleagues (Oslo Sports Trauma Research Centre, funded by FIFA's Medical Assessment and Research Centre); Petersen, Thorborg, Hölmich and colleagues (Copenhagen); the meta-analysis by Thorborg and colleagues
2015
Trackers allowed in matches
Clubs already put electronic performance and tracking systems (EPTS), such as GPS vests, local radio positioning and camera tracking, on players in training. On 28 February 2015, at its 129th annual meeting in Belfast, the IFAB (the body that writes football's Laws of the Game) approved EPTS in matches in principle, binding from 1 June 2015. Each league decides whether to allow them, and the data could not be used in the technical area during play. FIFA then set up a test program that checks each system against Vicon motion capture and Victoria University's Vision Kit tracking.
LessonFIFA says passing its test "does not constitute an endorsement of the quality of information generated by the device", so being allowed and being accurate are checked separately.
Credit
the IFAB; the FIFA Quality Programme for EPTS
2021
NFL settlement drops race-norming
Race-norming means adjusting a person's test scores by race. In the NFL's concussion settlement, the dementia tests assumed Black men had a lower cognitive baseline, which made it harder for Black former players to qualify for awards. After former players Kevin Henry and Najeh Davenport sued, the NFL and the players' lawyers filed a deal on 20 October 2021 to stop using race norms, and Judge Anita B. Brody approved it on 4 March 2022, letting players be retested or have their claims rescored. AP reported that the change could add $100 million or more to the NFL's costs.
Lessonneurologists built the race adjustment in the 1990s for clinical use, and experts quoted by AP said it was never meant to decide payouts in a court settlement.
Credit
Kevin Henry, Najeh Davenport and their lawyer Cyril V. Smith; Senior US District Judge Anita B. Brody
2024
Mouthguards trigger head checks
From 1 January 2024, World Rugby's head injury assessment (HIA) protocol for elite competitions used instrumented mouthguards, which contain an accelerometer and a gyroscope and alert the independent match-day doctor when a head acceleration passes a set limit. The first limits were 70 g for men and 55 g for women (g is the pull of gravity). After a 2023 women's tournament produced 30 alerts in 27 games, World Rugby raised them to 75 g and 65 g, and in March 2024 it changed the process again because in some stadiums alerts reached the pitch-side doctor too slowly. RNZ reported that every Super Rugby Pacific player wore one in 2024.
Lessonthe first limits for women rested on too little data and pulled too many players off, so they were revised within months.
Credit
World Rugby (chief medical officer Éanna Falvey), on advice from its independent Concussion Working Group
Watch for
Caps halve practice concussions
In 2022 the NFL required linemen, tight ends and linebackers to wear the Guardian Cap, a padded shell over the helmet, in preseason practice. A study of preseason practices from 2018 to 2023 found concussion rates in those positions then fell 54% in a simple comparison and 62% in a statistical model, but found no significant link for concussions involving the helmet shell, the part the cap covers. The authors work for Biocore, which takes consulting fees from the NFL; the NFL's chief medical officer is a co-author; and the NFL-backed Football Research Inc. paid for the study. An independent study from the same period measured 7,509 head impacts in 54 college players with instrumented mouthguards and found the cap had no effect on how hard or how often heads were hit.
Achieved whena group without NFL or Guardian Sports funding reports, in a peer-reviewed paper, fewer concussions in professional players who wear the cap than in comparable players in the same seasons, including for hits to the helmet shell.
Claimed by
Biocore researchers and the NFL's chief medical officer, American Journal of Sports Medicine, July 2025
AI tool predicts NFL injuries
The NFL says its Digital Athlete system, built with Amazon Web Services (AWS), takes video and sensor data from training, practice and games for all 32 teams and gives each team's medical staff injury-risk information on its players. The league says overall injuries have fallen since the tool arrived, and it used the tool to simulate 10,000 seasons when modeling a new kickoff rule. Tyler Williams, who runs health and performance for the Minnesota Vikings, told AP he could not be sure the tool directly caused the drop.
Achieved whenthe NFL or AWS publishes how well the tool's risk scores predicted injuries in seasons the model had not seen, and an outside group confirms those numbers or links the tool's use to lower injury rates.
Claimed by
the NFL and AWS, reported by AP, October 2025
Open
01
Open rugby head-contact labels
A public table of head-contact events in professional rugby union and rugby league, built from match and highlight videos that leagues post publicly. Each row holds a video link, start and end times, a contact type (head-to-head, head-to-shoulder, head-to-ground, other) and a coder ID. Publish the labels, a coding manual and a baseline detector's scores. Publish pointers to the videos rather than the files, and include no player names and no medical outcomes.
The briefCredit · unclaimed02
Injury rates from public reports
An open, versioned table of time-loss injuries (injuries that cost a player games) taken from public league injury reports, for at least three leagues, including one outside North America and one women's league, from 2020 on. Code each record to a body area using the Orchard Sports Injury Classification System, code only what the report states, and join each record to public playing time so that rates per 1,000 player-games can be calculated. Release stable player IDs instead of names, with the coding manual and the code.
The briefCredit · unclaimed
Open 01 · the brief
Open rugby head-contact labels
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of head-contact events in professional rugby union and rugby league, built from match and highlight videos that leagues post publicly. Each row holds a video link, start and end times, a contact type (head-to-head, head-to-shoulder, head-to-ground, other) and a coder ID. Publish the labels, a coding manual and a baseline detector's scores. Publish pointers to the videos rather than the files, and include no player names and no medical outcomes.
Why it is still open
In the most recent study found (Mohan and colleagues, Injury Prevention, 2025), a team trained a head-on-head detector on 1,011 professional rugby league clips, with one analyst coding them, and released no data. The authors say the only other way to measure head-impact exposure from video is manual analysis, which is "extremely labour intensive". American football already has open sets from the NFL's Kaggle competitions, and football (soccer) headers are labeled in SoccerNet's Ball Action Spotting set, but no open rugby set was found.
eprints.leedsbeckett.ac.ukkaggle.comraw.githubusercontent.com
Built with what exists
a video labeling tool such as Microsoft VoTT, which the 2025 study used; for the baseline, that study's recipe of a YOLOv7 head detector trained on the public CrowdHuman image set followed by a 3D convolutional network that classifies the impact.
First step
Write a one-page coding manual that reuses the 2025 study's definition of head-to-head contact, "resulting in a visible change in the trajectory of the head". Then label every tackle in one publicly posted full match, and have a second person label the same match to measure agreement.
Done looks like
At least 1,000 labeled tackle events from at least two competitions, including at least 100 head-on-head events (the 2025 study's power calculation needed at least 92). A second coder re-labels a random 20% with Cohen's kappa, a score for agreement beyond chance, of at least 0.80. A fixed split into training and test sets, and the baseline's sensitivity and specificity on the test set reported beside the 2025 figures (65% to 68% sensitivity, 84% specificity).
Open 02 · the brief
Injury rates from public reports
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open, versioned table of time-loss injuries (injuries that cost a player games) taken from public league injury reports, for at least three leagues, including one outside North America and one women's league, from 2020 on. Code each record to a body area using the Orchard Sports Injury Classification System, code only what the report states, and join each record to public playing time so that rates per 1,000 player-games can be calculated. Release stable player IDs instead of names, with the coding manual and the code.
Why it is still open
The closest open dataset (Bullock and colleagues, Scientific Reports, 2021) covers only four men's North American leagues from 2007 to December 2019, was scraped from a third-party website and is coded by body area only. Its authors write that it "did not include exposure or severity information" and that "Future research is needed to improve the open data extraction methods, access, and validation methods." The nflreadr package serves NFL weekly injury reports from 2009 on, but for one league only.
Built with what exists
nflreadr's load_injuries() for NFL reports; R with the rvest and xml2 packages, which Bullock and colleagues used, and their published extraction code as a starting point; the open Orchard classification.
First step
Load one NFL season with
load_injuries(), code 200 entries to Orchard body areas using a draft manual, and log every entry the manual cannot place.Done looks like
Records for at least 3 leagues across at least 5 seasons each, every row with a source link and date. A second coder re-codes a random 10% of body-area codes with Cohen's kappa of at least 0.80. Playing time joined for at least 95% of player-seasons. A note comparing one injury count with an independent public source, as the 2021 study did.